Previously developed TLC methods for the determination of elemental sulfur were normal-phase separations with unmodified silica as the stationary phase; in some cases, mixtures of silica with magnesium silicate were used. Solvents were generally very apolar, comprising petroleum ether, n-heptane, hexanes, 2,2-dimethylbutane, and tetrachloromethane, in rare cases with polar modifiers [11,12,13,14]. We therefore based our method development on silica TLC plates and used the well-defined solvents n-heptane, n-hexane, cyclo-hexane (cHx), and toluene. Only neat solvents were considered, to avoid variation during the preparation of solvent mixtures in routine use. Petroleum ether, which is a mixture of hydrocarbons defined only by a boiling range, and the less common 2,2-dimethylbutane were also rejected as mobile phases. During this initial stage, TLC plates with fluorescence indicators were used. These allow the direct detection of elemental sulfur, which absorbs ultraviolet (UV) light. For all tested solvents, very high RF values were observed for sulfur (0.94 for cHx, 0.88 for n-heptane) which eluted as a single peak without the undesired separation of the allomorphs that can be observed in other chromatographic techniques (Fig. 1, left, shows exemplary chromatograms). The samples were applied without pretreatment, and it was found that sulfur eluted as predicted while other liquor components did not at all migrate in these very apolar solvents. Therefore, the short developing distance of 30 mm was chosen to save time, and sample pretreatment was found to not be required. As expected, chromatography was markedly slower for cHx: it took 200 s (3.33 min) to cover a developing distance of 30 mm compared with 86 s (1.43 min) for n-hexane. This seeming lack of speed is a minor issue in the light of the already very short effective developing times per sample: 8.3 s compared with 3.6 s. The shape and intensity of the obtained peaks was most defined when cHx was used for development. When n-hexane or toluene was used, spots were slightly paler and more diffuse and became indistinct with n-heptane. cyclo-hexane was therefore selected as eluent. TLC plates were selected as stationary phase to save costs, since the clear separation did not require a higher separation efficiency.

Left: typical chromatograms of sulfur standards (applied amount indicated) and black liquor samples. The chromatograms recorded by densitometry at 285 nm are overlaid in purple. Right: UV/vis spectrum of elemental sulfur recorded off the TLC plate

For the preparation of sulfur standard solutions, a range of apolar solvents is presented in the literature, including tetrachloromethane, chloroform, carbon disulfide, acetone, and hexane [11,12,13,14]. Also here, cHx was selected, since it gave the clearest and most intense bands and allowed to dissolve sulfur at concentrations higher than 1 mg/mL. Toluene even showed detrimental band broadening during application. In contrast to other separation methods, it is not a problem in TLC that the sample solvent–water–is immiscible with the solvent of the standards and the eluent–cyclo-hexane–since all solvents are evaporated before chromatography.

Sulfur absorbs light in the UV range. Consequently, we used plates with a UV-sensitive fluorescence indicator to detect it without derivatization during method development. Silver and rhodamine were initially tested as staining reagents. While both yielded reasonable first results, these options were not pursued further for the sake of simplicity, robustness, and resource efficiency. Instead, we employed scanning densitometry for the direct detection of sulfur at a specific wavelength. The absorption maximum at 285 nm was confirmed by a spectral scan (see Fig. 1, right), and this wavelength was used for detection by scanning densitometry in all further analyses [10, 11, 13, 14]. With the direct detection by scanning densitometry, a fluorescence indicator in the stationary phase was no longer necessary, therefore running costs for the analysis were reduced by switching to aluminum-backed TLC-grade plates without fluorescence indicator. On these plates, peak shape and RF of sulfur, as well as the immobility of the other liquor components remained as previously established during method development. Due to the short developing distance of 30 mm, the plates can be developed from either side or be trimmed after each separation to be used for two or three separations to increase resource efficiency. Care must be taken that the plates are cleaned by washing or pre-development with methanol before chromatography. Otherwise, contaminants that were adsorbed to the plate during storage and that elute in the solvent front can overlap with sulfur.

The application of a defined sample volume to the plate is fundamental for quantification and can be done by two methods: spraying or spotting. Spraying is generally preferred, as it gives sharp, homogeneous lines, which is beneficial for both resolution and quantitative evaluation. To our surprise, no sulfur was detected in liquor samples that were sprayed onto the plate. We did not investigate the cause for this, but suspect that the polymeric liquor components encapsulated sulfur in a process comparable to spray drying. Samples and standards were therefore spotted onto the plate. As an advantage, more samples can be fit onto one plate when spotting—24 instead of 15 spots at the chosen spraying settings—which reduced the effective analysis time per sample by 38%. Reproducibility of spotting was found to depend on the spotting speed: higher flow during spot application (200 and 250 nL/s) resulted in a smaller variation of peak area compared to lower flow (50 and 100 nL/s). Since this directly affects the precision of the calibration, the fastest setting was selected for sample application. Care must be taken to properly rinse the capillary that is used for sample application when changing from samples to standards, since the solvents of samples and standards—highly alkaline solutions and cyclo-hexane, respectively—are immiscible. For an automated system as the one used, it is recommended to rinse consecutively with water, 2-propanol, or acetone, and then with cyclo-hexane when changing from alkaline samples to sulfur standards. Although this adds to the time required for a full analysis, it can only be avoided by the less-precise manual sample application.

Quantification was based on peak areas, since spotting liquor samples resulted in tailing peaks with varying height. For validation, three separate sets of standards in a range from 10 to 3000 µg/mL were analyzed in quadruplicate on three separate plates. First, a suitable calibration function was identified. Four established calibration models were applied to the validation measurements: a second order polynomial, a linear correlation of the square root of the peak area against the logarithm of the concentration, and Michaelis–Menten functions both with and without axis intercept [8, 19,20,21]. Only the Michaelis–Menten function with axis intercept passed the F-test at a confidence level of 0.95, while the Michaelis–Menten function without axis intercept narrowly failed the test. The polynomial could not model the lowest concentrations (10 and 30 µg/mL) and had its maximum within the calibration range at 2300 µg/mL. The logarithmic fitting function clearly failed the F-test. The selected calibration function was therefore a Michaelis–Menten function with axis intercept (see Fig. 2, left).

Left: typical calibration curve for sulfur at 285 nm as recorded during method validation. Right: accuracy profile of the developed method. Circles denote the bias for each individual measurement. The black solid line indicates the mean bias, while the dashed gray lines indicate the upper and lower limits of the β-expectation interval. Lines at 0 and ±20% bias are provided for orientation

For validation, three separate stock solutions and dilution series were prepared that covered the range from 10 to 3000 µg/mL (20–6000 ng on plate). These solutions were each analyzed in quadruplicate; this allows taking into account the variations both within a plate and between plates. With this approach, it is possible to capture both accuracy (as mean bias, the average deviation from the expected value) and precision (as β-expectation interval, the interval that is expected to cover 80% of the future measurements) at each calibration level, and the cumulative influences of all analytical steps from the preparation of standards onwards are included (see Tables S1 and S2 in the Supplementary Material for the actual numbers). It became apparent (Fig. 2, right) that for the lowest level (10 µg/mL, 20 ng), sulfur could only be detected for some of the samples, and that at the highest level (3000 µg/mL, 6000 ng) the calibration yielded meaningless results. The validation was therefore limited to the range of 60–2000 ng on plate. At the lowest evaluated level, sulfur was reliably detected. The average bias of about ‒5% indicates an accurate evaluation, even though with limited precision. Therefore, 60 ng on plate can be assumed as the limit of quantification; the corresponding limit of detection would be placed at 20 ng, which is supported by the erratic detection of sulfur at 10 ng. The limit of detection and the limit of quantification are therefore between these values. The average bias then reaches about +20% at 200 ng and falls to −10% at the highest concentration. This behavior indicates that the calibration function, while it was the best of the investigated, passing a formal test and being commonly used, is still not optimal and introduces some systematic deviation. That the calibration points for 200 ng are slightly above the calibration curve can already be seen in Fig. 2, left. The determined β-expectation interval—it describes uncertainty, as 80% of all future measurements should fall into this interval—is below 20%, which is an acceptable value for this type of analysis. Based on this information, the method can be considered fit for the intended purpose.

The developed method was found to be robust against most variations in separation conditions. The material of the stationary phase did not influence the elution of sulfur, as long as it was unmodified silica. Since resolution is not critical, TLC-grade silica can be chosen over its high-performance counterpart. Plates with fluorescence indicators can be used to detect sulfur directly at an increased limit of detection (LOD) of about 60 ng. However, prevalidation indicated a decreased sensitivity for densitometric quantification for plates with fluorescence indicators; therefore, plates without indicators were used. The solvent system consists of a single component, which avoids variations in composition caused by errors during preparation or lengthy storage. Based on the experiences made during method development, several apolar hydrocarbons besides cyclo-hexane (n-hexane, n-heptane, toluene) should give a useful separation. However, these solvents might reduce the quantitative performance. Separations were performed without chamber saturation, and identical results were achieved in regular (twin-trough) and horizontal TLC chambers. In agreement with theory, chamber saturation reduced the RF value of sulfur from 0.94 to 0.56; at the same time, the development was accelerated [22]. The peaks obtained in saturated chambers were markedly broader than in unsaturated chambers, which can be expected to affect the LOD negatively. Relative humidity did not have an effect in the tested range (0–75%). It was essential to apply the samples by spotting, as spraying completely prevented octa-sulfur to elute from the application line. Precision was influenced by dosing speed, with faster speeds resulting in more reproducible peak areas and shapes. Prewashing the plates with methanol was crucial for quantitative analysis, as it removed compounds adsorbed to the plate, which might elute with the solvent front and overlap with the sulfur peak.

Resource consumption of this method is comparatively low due to the parallel analysis of several samples on each plate. At least 24 samples can be applied to each 20-cm-wide plate without interfering with each other. To develop a single plate, 10 mL of eluent and 103 min are required: sample application (84 min), development and drying (15 min), and detection (3 min). Thus, the analysis of each sample requires 0.42 mL of non-halogenated solvent and 4.3 min. Automatic sample application takes the most time, which includes an exchange of the machine’s rinsing solvent from alkaline buffer over water and 2-propanol to cyclo-hexane. Some minutes might be saved by a manual application. Then, a solvent exchange is not necessary, but one must be aware that this will impair quantification. Given the short developing distance, a 10-cm-high plate can be used for up to three separate analyses (72 samples), if the used area of the plate is simply cut off after each analysis. No derivatization is necessary to visualize the target compound, if plates with fluorescence indicators are used.

Finally, the developed method was put into use. Nine black liquors were sampled from industrial Kraft processes. First, ICP‒OES was used to determine the total amount of sulfur in the samples, which captures not only elemental sulfur but also different sulfur salts from pulping and bleaching, organic sulfur compounds, such as dimethyl disulfide as well as sulfur bound to the released lignin. For this measurement, all organically bound sulfur is released during sample pretreatment by oxidative digestion. In these black liquors, the total sulfur content ranged from 8.6 to 13.3 g/L. The liquor samples were then analysed without pretreatment with the developed TLC method (see Fig. 3). Samples were analyzed in octuplicate on four different plates. The sample volume was increased to 6 µL to ensure that the applied amount of elemental sulfur was within the calibrated range. The observed concentrations in the black liquor were in the range of approximately 40–55 mg/L; only one sample ranged clearly lower at 27 mg/L (see Table 1). The uncertainty of the method was slightly larger than expected from validation, with a range from 13% to 45% and an average of 30%. In relation to the total mass of sulfur, elemental sulfur turned out to be a small fraction: typically, it made up 0.3–0.5% of the sulfur in black liquor, again with the exception of one single sample that only reached 0.24%. Generally, the portions of elemental sulfur were found to be around 0.4%, ranging from 0.24% to 0.48%.

Analysis of black liquors according to the established protocol. A plate with fluorescence indicator was used to visualize the sulfur for this graph. Standards are in lanes 2, 4, 7, 10, 13, 16, and 20